bounds values set wrongly after calling .select on Image object

[RafalSkolasinski]

When .select method is called on Image object it looks like bounds of newly created image are set to the input of .select method. I found that in following case:

import numpy as np
import holoviews
from holoviews import Image, Points
holoviews.notebook_extension()

side = 5
region=(-side,-side,side,side)

def coords(bounds,samples):
    l,b,r,t=bounds
    hc=side/samples
    return np.meshgrid(np.linspace(l+hc,r-hc,samples),
                       np.linspace(b+hc,t-hc,samples))

x,y = coords(region, 5)
z = np.sin(x**2+y**2)

img1 = Image(z, group="Function", label="Sine", bounds=region) 
img2 = img1.clone(extents=(-2, -2, 2, 2))
img3 = img1.select(x=(-2,2), y=(-2,2))

>>> print(img1[0.95, 0], img2[0.95, 0], img3[0.95, 0])
0.0 0.0 -0.756802495308

%%opts Image [situate_axes=True] {+axiswise}
img1*Points([(0.9,0)]) + img2*Points([(0.9,0)]) + img3*Points([(0.9,0)])

it leads to confusion when accessing value at the same coordinates of the initial and new object.
Maybe bounds of a new object should be selected to the edge of edge pixels that got selected?

[philippjfr]

Thanks @RafalSkolasinski. This a pretty major and annoying bug and will have to be addressed.

[jlstevens]

I've assigned the milestone to 1.6.0 as I agree this is an important bug to address. I would consider moving it to a minor, post 1.5 release if we decide we are going to have one of those (seems likely!).

[RafalSkolasinski]

Hmm... maybe it would be good idea to make

[situate_axes=True]

a default setting?

[philippjfr]

This turned out to be a pretty important bug which actually caused some distortions in a plot in my thesis. I've finally figured it out in #994, which will hopefully make it into the 1.7 release.

[philippjfr]

This has now finally been finally been fixed in #994, which was just merged.

>>> print(img1[0.95, 0], img2[0.95, 0], img3[0.95, 0])
(0.0, 0.0, 0.0)

%%opts Image [situate_axes=True] {+axiswise}
img1*Points([(0.9,0)]) + img2*Points([(0.9,0)]) + img3*Points([(0.9,0)])